Integration of experimental, microstructural, and machine learning-based evaluation of recycled aggregate concrete incorporating silica fume and nano silica

Abstract The sustainable alternatives to conventional construction materials are beneficial for infrastructure development. Using recycled coarse aggregate (RA) in concrete makes construction more sustainable.This study investigated the experimental, microstructural performance and machine learning examination of RAC mixes, using 0–65% RA replacement, where RA12.5% exhibited optimum performance and was further modified using Silica fume (SF) at 5–10% and Nano silica (NS) at 1–2%, and assessed the mechanical properties, Ultrasonic pulse velocity (UPV), at 7, 28, and 90 days curing. The results show that as %RA increases reduces the mechanical properties and UPV due to higher porosity and weaker interfacial transition zones (ITZs). The RA12.5% + NS1% mix achieved the highest compressive strength (62.28 MPa), split tensile strength (4.69 MPa), and flexural strength (7.90 MPa) at 28 days, while the RA12.5% + SF5% mix recorded the highest UPV (5500 m/s). Both SF and NS increased the concrete density, reduced porosity, and promoted C–S–H gel formation. EDS analyses confirmed improved hydration characteristics with lower Ca/Si ratios and denser microstructures in SF and NS incorporated mixes. Furthermore, machine learning models were employed for performance prediction. Extra Trees and Gradient Boosting demonstrated superior prediction accuracy for strength prediction with testing R 2 values of 0.932 and 0.930, respectively, while kNN achieved the highest UPV prediction accuracy with R 2 = 0.980. The results confirm that optimized incorporation of SF and NS effectively enhances the performance and sustainability of RAC, while ensemble learning algorithms provide reliable predictive capability for concrete property assessment.